Open Source Initiative
The Open Source Initiative (OSI) is a California nonprofit that maintains the Open Source Definition, the ten-point standard used to decide whether a software license counts as open source, and that reviews and approves licenses against it.[1][2][3] Eric Raymond and Bruce Perens founded it in February 1998, weeks after the term "open source" was coined, and it adapted the definition from the Debian Free Software Guidelines.[1] It is organized as a California public benefit corporation with 501(c)(3) tax-exempt status and is run by a volunteer board and a small staff.[3][4]
In artificial intelligence, OSI is best known for the Open Source AI Definition (OSAID). Version 1.0 was announced on October 28, 2024, after research that began in 2022 and public drafting that began in 2023, and it remained the current version on OSI's website as of November 2026.[5][6][7] The definition says what a machine learning system must make available to be called open-source AI: the trained parameters, the complete training and inference code, and detailed information about the training data, though not necessarily the data itself.[7] That last choice drew sustained criticism from parts of the free software community, and OSI's public statements that Meta's Llama models are not open source put it in open dispute with the company.[8][9][10]
OSI has no power to enforce either definition. Its influence comes from other parties, including governments and standards bodies, choosing to refer to its license list and its definitions.[1][10] As of November 2026 its executive director is Duane O'Brien, who took up the post on April 13, 2026.[11][12]
History
Founding and the "open source" label
OSI's own history dates the label "open source" to a strategy session held in Palo Alto, California, on February 3, 1998, shortly after Netscape announced that it would release its browser's source code. The participants wanted a single term for the collaborative development model that would be easier to present to businesses than "free software", which they regarded as philosophically and politically focused. Christine Peterson suggested "open source". Those present included Eric Raymond and Michael Tiemann, both later presidents of OSI, along with Jon "maddog" Hall, Larry Augustin and Sam Ockman.[1]
Raymond and Perens founded the organization in late February 1998, with Raymond as its first president and Perens as vice-president. The initial board included Brian Behlendorf, Ian Murdock, Russ Nelson and Chip Salzenberg.[1]
One of the first tasks was to write the Open Source Definition (OSD). Perens had drafted the Debian Free Software Guidelines, which Debian's developers adopted as formal policy in 1997, and the OSD was produced in February 1998 by revising those guidelines and removing the Debian-specific references. OSI published its first formal list of approved licenses by October 1999.[1]
From volunteer board to staffed organization
OSI received recognition as a 501(c)(3) nonprofit from the US Internal Revenue Service in 2003. In 2011 and 2012 it added an affiliates program for other open-source organizations, board seats nominated by those affiliates, and an individual membership program.[1]
In September 2021 OSI appointed Stefano Maffulli as its first executive director, describing the hire as part of a move, begun by the board in 2020, to become a professionally managed organization. The role of board president was replaced by a board chair at the same time.[13] OSI later credited Maffulli with turning a largely volunteer-run group into a professionally managed nonprofit with full-time staff.[14]
The Open Source Definition and license review
The OSD (version 1.9) sets conditions on the terms under which software is distributed. Its opening line says that open source "doesn't just mean access to the source code".[2] The ten criteria are:
| No. | Criterion |
|---|---|
| 1 | Free redistribution |
| 2 | Source code |
| 3 | Derived works |
| 4 | Integrity of the author's source code |
| 5 | No discrimination against persons or groups |
| 6 | No discrimination against fields of endeavor |
| 7 | Distribution of license |
| 8 | License must not be specific to a product |
| 9 | License must not restrict other software |
| 10 | License must be technology-neutral |
Criteria 5 and 6 are the ones OSI has invoked against AI model licenses. Criterion 6 reads: "The license must not restrict anyone from making use of the program in a specific field of endeavor."[2] On OSI's reading, a license that limits commercial use for some users, or that attaches an acceptable use policy, fails on those grounds whatever else it permits.[9][15] Clause 10 was added in 2004; OSI describes the text as otherwise stable since 1998 apart from minor clarifications.[1]
Licenses are approved through a public process. A submitter posts the license to OSI's license-review mailing list, where anyone may comment. The normal decision date is 60 days after submission. OSI's License Committee judges whether the discussion has reached consensus, its chair makes a recommendation, and the board votes. A "legacy" license, one already used for at least five years by more than twenty projects from unrelated organizations, is assessed as written; a new license must also explain what gap it fills and how it compares with similar approved licenses.[16] By page views on OSI's site in 2025, the most-read approved licenses were the MIT License, the Apache License 2.0 and the two BSD licenses.[17]
In 2025 OSI approved the OSC License and a revision of the Common Development and Distribution License, and its licensing community discussed the ModelGo licenses, a family written for AI models.[17] In November 2026 OSI reported that two AI-specific licenses, ModelGo and OpenMDW, were awaiting its approval.[45] Since May 2026 the License Committee has been able to reject submissions itself under certain conditions; in September it reported four rejections for poor drafting.[18]
Governance and funding
Board
The board of directors is the organization's ultimate authority: it approves the budget, sets strategy and oversees the executive director. Directors are volunteers and must sign a board member agreement.[3] For over a decade the board combined seats filled after votes by individual members and by affiliate organizations with seats appointed by the board itself. Those votes were always advisory; the sitting board made the formal appointments.[3][19]
After a disputed election in 2025 (see below), OSI did not hold its spring 2026 election. Interim executive director Deborah Bryant wrote in January 2026 that "the experiment has not produced the results we hoped for" and that the process had too often caused confusion and diminished trust.[19] A Governance Working Group chaired by director Ruth Suehle reviewed how directors are chosen.[20] On September 24, 2026 the board adopted its recommendations:[18]
- Votes by individual supporters and affiliates are treated as candidate ranking that informs board appointments, not as binding elections.
- The target board has 13 seats: four ranked by individual supporters, four ranked by affiliates and five appointed directly by the board.
- All seats carry three-year terms, and expiring terms were extended into 2027 to bridge the change.
- Staff were authorized to draft bylaw amendments with legal counsel, including a proposed cap of two seated directors from any single corporate group.
As of November 17, 2026, OSI's board page lists nine directors: Anne-Marie Scott (treasurer), Carlo Piana, Catharina Maracke, Chris Aniszczyk, Gaël Blondelle (secretary), McCoy Smith (chair of the License Committee), Ruth Suehle (vice chair), Sayeed Choudhury (vice secretary) and Thierry Carrez. The page shows no director with the title of chair.[3] Tracy Hinds, named as board chair in OSI announcements in September 2025 and April 2026, presided over the September 24, 2026 meeting. Her term ended on September 30, 2026, and O'Brien's October 2026 newsletter thanked her as outgoing chair; OSI had not named a successor on its board page as of November 17, 2026.[14][11][18][43]
Executive leadership
| Period | Person | Role |
|---|---|---|
| 1998 | Eric Raymond | First president [1] |
| Until September 2021 | Deb Nicholson | Interim general manager [13] |
| September 2021 to October 2025 | Stefano Maffulli | First executive director [13][14] |
| September 2025 to April 2026 | Deborah Bryant | Interim executive director [14][11] |
| April 13, 2026 onward | Duane O'Brien | Executive director [11][12] |
OSI announced on September 15, 2025 that Maffulli would leave in October to work on open-source AI and data governance, and that Bryant, a former board member and former US policy director for OSI, had been appointed interim executive director.[14] O'Brien's appointment was announced on April 15, 2026. He had run Capital One's open source program office and earlier built the one at Indeed.[12] In his first post as director he wrote that work on the OSAID should continue in the open, "grounded in community input, transparent process, and a commitment to definitions that are practical, durable, and widely understood".[4]
Finances and sponsors
OSI's income comes from corporate sponsorship, contributions, grants and membership dues. Its 2025 annual report shows a sharp fall in revenue and a board-authorized draw on reserves:[17]
| Item (US$ thousands) | 2024 | 2025 |
|---|---|---|
| Sponsorships | 520 | 357 |
| Contributions | 227 | 204 |
| Grants | 249 | 50 |
| Membership dues | 26 | 24 |
| Event and other revenue | 69 | 32 |
| Total revenue | 1,091 | 667 |
| Total expenses | 1,035 | 1,129 |
| Net income | 56 | -462 |
| Cash at year end | 970 | 432 |
The report attributes the drop to weaker technology sponsorship, the executive director transition and a decision to pursue little grant funding in 2025. Of $631,025 in program spending, 59.9% went to policy and standards work, 15.5% to the ClearlyDefined license-metadata project, and 9.2% ($57,924) to AI.[17]
As of November 2026 the sponsors page lists Automattic at the top tier, followed by MongoDB, Cisco, Amazon Web Services and Google Open Source, then Bloomberg, Microsoft and Red Hat, with GitHub among those in the next tier. The page states that OSI does not endorse its sponsors.[21] TechCrunch reported in October 2024 that OSI's funders then included Meta, Amazon, Google, Microsoft, Cisco, Intel and Salesforce, and that OSI had obtained a grant from the Alfred P. Sloan Foundation to reduce its reliance on technology companies.[10]
Work on artificial intelligence
From Deep Dive: AI to OSAID 1.0
OSI began studying AI in July 2022 with a podcast and panel series called Deep Dive: AI. Its later report to the board said it became apparent in 2022 that, because of technical, semantic and economic differences between software and AI systems, a simple translation of the OSD would not carry the open-source freedoms over to AI.[6] In June 2023 Maffulli announced that OSI would write a definition, objecting to large language models and tools being promoted as open source while carrying restrictions the OSD forbids.[22]
OSI at first floated a slower process modeled on the drafting of version 3 of the GNU General Public License, running through 2024 and concluding in 2025. Its report says community partners were concerned about the timeline, fearing that "Open Source AI" would become a generic term and that European Union regulators would write a definition without community input. The board set a deadline of October 2024.[6]
| Date | Step |
|---|---|
| July to October 2022 | Deep Dive: AI podcasts and panels [6] |
| June 21, 2023 | Kickoff meeting at Mozilla's San Francisco headquarters [22] |
| Late 2023 | Workshops in Raleigh, Monterey and Addis Ababa draft the four freedoms [6] |
| February to March 2024 | Working groups examine Llama 2, BLOOM, OpenCV and Pythia; draft 0.0.6, dated March 9 in OSI's drafts index, incorporates their recommendations [6] |
| Mid-2024 | Volunteers review 13 systems described by their makers as open, using versions 0.0.6 to 0.0.8 [6] |
| August 22, 2024 | Version 0.0.9 [22] |
| October 2, 2024 | Release candidate 1 [23] |
| October 21, 2024 | Release candidate 2, whose text became version 1.0 [47] |
| October 10-11, 2024 | Paris workshop on training data, 18 experts from 15 countries [6] |
| October 28, 2024 | Version 1.0 announced at All Things Open in Raleigh, North Carolina [5] |
OSI called the method "co-design" and brought in the firm Do Big Good to manage it. The report counts more than 50 volunteers from about 30 countries and says OSI sought a Sloan Foundation grant for global outreach after hearing that a process held only in person shut out people who could not travel to the workshops.[6] LWN reported that the board was due to vote on the text on October 27, 2024, the day before publication.[22] Carlo Piana, then board chair, called the process "well-developed, thorough, inclusive and fair".[5]
What the definition requires
The detailed criteria are covered in the separate article on the definition. In outline, OSAID 1.0 says an open-source AI system must grant the freedoms to use, study, modify and share it, and that exercising them requires access to the preferred form for making modifications, which has three parts:[7]
- Data information: "Sufficiently detailed information about the data used to train the system so that a skilled person can build a substantially equivalent system."
- Code: the complete source code used to train and run the system, under OSI-approved licenses.
- Parameters: the model weights and other settings, under OSI-approved terms.
The definition does not require release of the training data itself. OSI's FAQ sorts data into four classes (open, public, obtainable from third parties, and unshareable): open data must be shared, and the other classes must be listed or described in detail.[24][7] The text permits conditions that require modified versions to be released under the same terms, and it leaves open what legal instrument makes parameters free, saying this should "become clearer over time" as the legal system addresses such systems.[7] TechCrunch noted at launch that version 1.0 does not settle whether a copyright license on a model would be enough to satisfy it.[10]
Model weights and "open weights"
OSI applies the same requirements to a whole system, a model, or a set of weights. Under the OSAID, weights described as open source must come with the data information and code used to produce them.[7] A model whose weights are published under an approved license without that material is, in OSI's vocabulary, open weights and not open-source AI.[25]
OSI set this out in a January 2025 explainer. It defines open weights as "the final weights and biases of a trained neural network", credits licensing specialist Heather Meeker's 2023 Open Weights Definition with formalizing the idea, and describes such releases as more transparent than proprietary systems but lacking the training code and data documentation needed to audit or replicate a model.[25] In September 2025 OSI said that many people still used "Open Source AI" to mean open weights only and that it continued to push back.[26] An OSI post in August 2026 noted that half of the ten most-used open models in a Mozilla report carried an OSI-approved license, and added that this did not mean they met the OSAID.[27]
Positions on specific models and licenses
OSI's dispute with Meta predates the AI definition. In July 2023, shortly after Llama 2 was released, Maffulli wrote that its license was "very plainly not" an open-source license, because it limits commercial use by some users and restricts use for certain purposes through an acceptable use policy, contrary to OSD criteria 5 and 6. He said OSI had asked Meta to correct its description.[9] In February 2025 OSI repeated the charge for the Llama 3 series, saying the newer community licenses added restrictions, including one excluding people in the European Union from using a model, and asking readers to call on Mark Zuckerberg and Yann LeCun to change the license.[15]
Meta rejected the definition when it was published. A spokesperson told TechCrunch: "There is no single open source AI definition", and said the Llama license and acceptable use policy "help keep people safe". The spokesperson pointed to other efforts to define the term, including suggested definitions from the Linux Foundation and the Free Software Foundation's criteria. Maffulli told the same outlet that Google and Microsoft had agreed, after discussions with OSI, to stop using "open source" for models that are not fully open, and that Meta had not.[10]
During drafting, volunteers tested the definition against existing systems. OSI's FAQ reports the results, with the caveat that "they are not certifications of any kind":[24]
| Outcome | Systems |
|---|---|
| Passed | Pythia (EleutherAI), OLMo (Ai2), Amber and CrystalCoder (LLM360), T5 (Google) |
| Would probably pass with different legal terms | BLOOM (BigScience), StarCoder2 (BigCode), Falcon (TII) |
| Did not pass | Llama 2 (Meta), Grok (X/Twitter), Phi-2 (Microsoft), Mixtral (Mistral AI) |
Most reviewers could not find the documentation needed to reach a conclusion. OSI took this as a reason to go on approving licenses and legal terms, as it does for software, and not to certify individual AI systems.[6] The open-weight license comparison covers how current model licenses fare against the OSD.
Reception of the Open Source AI Definition
Endorsements
By December 2024 OSI counted endorsements from more than 20 organizations, among them the Mozilla Foundation, EleutherAI, the Common Crawl Foundation and the Eclipse Foundation, and from more than 100 individuals.[28] In OSI's launch announcement, Ayah Bdeir of Mozilla said the training-data approach "may not be a perfect place to be" but that insisting on a standard no model builder would meet "could end up backfiring", and Stella Biderman of EleutherAI called the definition "a necessary step".[5] Maffulli also pointed to a comment from Carnegie Mellon University as support for OSI's view of what the preferred form for modifying a model is.[22]
Criticism over training data
The main objection is that a system can qualify without its training data being available. Before release, julia ferraioli wrote that without the data the definition guarantees only the ability to use and distribute a system, plus fine-tuning and transfer learning on top of it. Tom Callaway argued that open data should be mandatory even if that meant fewer qualifying systems at first.[22] Luca Antiga, chief technology officer of Lightning AI, said that by not dealing with the licensing of training data OSI was "leaving a gaping hole".[10]
Three days after release, Bradley Kuhn of the Software Freedom Conservancy wrote that "the OSAID fails to require reproducibility by the public of the scientific process of building these systems". He argued that OSI had refused a stricter rule because it did not want a definition that no existing system could meet, that this compromise amplified the role of parties that would benefit financially from having their systems declared open source, and that users and content creators had been left out. He said the text should be reissued as recommendations, and announced that he would run for OSI's board to seek its repeal.[8]
The Free Software Foundation said on October 22, 2024 that it was preparing its own criteria for free machine learning applications, under which the software, the raw training data and the associated scripts must all grant users the four freedoms. The FSF statement did not mention OSI or its definition.[29] OSI's February 2025 post on Llama cited the foundation's assessment of the Llama 3.1 license with approval.[15]
Debian developers, whose guidelines the OSD was derived from, debated a stricter line. In April 2025 Mo Zhou proposed a general resolution stating that AI models released under a compatible license but without their original training data or training program are not compliant with the Debian Free Software Guidelines and cannot enter the distribution's main archive. LWN noted that passage would put Debian in contrast with the OSAID. The proposal gathered enough seconds to proceed and was later withdrawn.[30][31]
Other objections
Some critics doubt that "open source" should be extended to AI at all. Stephen O'Grady of the analyst firm RedMonk wrote that the OSD succeeded partly because it removes nuance, while the OSAID's compromises need explanation, and he cited the political adage "If you're explaining, you're losing."[22] Amanda Brock, chief executive of OpenUK and a former OSI director, argued nearly a year after release that openness can only be defined per component (software, data, models, weights) and that adding a second definition had started to undermine the first.[32]
OSI co-founder Bruce Perens, in the same September 2025 article, said the definition had not reduced openwashing and that "Most AI vendors have not accepted OSI as an authority, with statements by Meta being the most glaring example."[32]
OSI's own report to its board records comments that the co-design process "was not democratic and ultimately unfair", that voting was the wrong method and that the selection of volunteers was biased. It also acknowledges a discrepancy in how one working group's votes were tabulated, which it says did not change the result once corrected.[6]
OSI's response
OSI's position is that training data is valuable for studying a system but is not part of the preferred form for modifying an existing one, and that some data, such as personally identifiable or patient data, cannot be shared.[22][24] Its FAQ says a full-data requirement would confine open-source AI to systems trainable only on open data, and "That niche would be tiny".[24] It has also said that reproducibility was never the aim of open source, for software or for AI, and that others can add requirements on top of its definition.[23] In February 2025 it accused unnamed critics of the OSAID of "supporting Meta's open washing efforts" by entangling their objections with the Llama dispute.[15]
OSI has presented 1.0 as a starting point. The board voted in late 2024 to set up an AI committee to work out how the text would be updated, and the report to the board said that committee would suggest reviews of the text at quarterly intervals. The same report said OSI had started conversations with Hugging Face, Carnegie Mellon, Mozilla and Google about how the definition is applied.[33][6] No revised version had been published as of November 2026.[7]
The 2025 board election
The dispute carried into OSI's governance. Kuhn and Richard Fontana ran in the March 2025 board election on a platform of repealing the OSAID and removing a clause in the board agreement that requires directors to "support publicly all Board decisions, especially those that do not have unanimous consent". Debian developer Luke Faraone, running on an aligned ticket, was kept off the ballot for missing a nomination deadline whose time of day, LWN found, had been poorly advertised.[34]
OSI excluded three candidates from the final tally, two because they had not signed the current board agreement and one for returning it after the deadline.[35] According to LWN, the requirement that candidates sign the agreement was added after voting ended and before the results were tallied. Kuhn and Fontana returned altered versions and objected that they had been told the agreement applied to seated directors, not to candidates. Maffulli said the excluded candidates had "excluded themselves". The board confirmed Carlo Piana, seated Ruth Suehle and McCoy Smith, and ordered a retrospective.[34][35] When the board paused elections in January 2026, Bryant cited community concerns about that election cycle.[19]
Policy work
Policy and standards work was OSI's largest program by spending in 2025. The annual report ties much of that to standards for the EU Cyber Resilience Act; the work related to AI regulation is summarized here. OSI also runs the Open Policy Alliance, a coalition of nonprofits that comments on public policy affecting open source.[17]
European Union
The EU AI Act contains exemptions for open-source AI, and OSI has said that lawmakers struggled to define which systems qualify.[26] OSI took part in drafting the GPAI Code of Practice from August 2024. It objected that early drafts required acceptable use policies and bans on certain uses, which conflict with OSD criterion 6, and it wrote to the drafting chairs with other organizations. The third draft, published March 11, 2025, made such policies optional and exempted open-source AI from prohibiting downstream uses.[36]
On April 10, 2025 a cross-party group of Members of the European Parliament led by Birgit Sippel and Markéta Gregorová wrote to Commission Executive Vice-President Henna Virkkunen. The letter said that a few very large companies were calling their AI open source while imposing usage restrictions, named Meta's Llama as an example, cited OSI's 2023 statement, and urged the Commission to clarify that such systems "cannot be considered Open Source for the purposes of the AI Act".[37] OSI's annual report says it supplied technical expertise to the parliamentarians and separately sent the Commission formal feedback on the definition of open-source AI.[17] Guidelines the Commission published on July 18, 2025 set out its own conditions for the exemption: the license must allow use, access, modification and redistribution, and may not impose non-commercial limits or user-size thresholds.[38]
In February 2025 OSI and the think tank Open Future published a white paper, "Data Governance in Open Source AI", based on the Paris workshop.[39] A three-day online conference on the same subject in October 2025 was Maffulli's last project for OSI as director.[40][17]
G7 Vision on AI openness
On May 29, 2026 the G7's digital and technology ministers, meeting in Paris under the French presidency, approved a non-binding "Vision on AI Openness Opportunities and Shared Language". OSI acted as a knowledge partner for three months, helping to draft the text and taking part in the negotiations, and the ministerial declaration acknowledged its contribution by name.[41]
The Vision states that "The openness of an AI is not binary. It exists on a spectrum", and it sets four labels for degrees of openness:[41][44]
| Label | Meaning in the G7 Vision |
|---|---|
| Weights Available AI | Weights and deployment code released free of charge under a license that restricts use (commercial, geographical or use-case limits) |
| Open Weights AI | Weights and deployment code released free of charge under an open-source license |
| Open Source AI | Broadly follows the OSAID, but training data may be withheld only where publishing it is legally or technically impossible |
| Open Source AI with Open Data | Weights, deployment code, training code and full training data all released free of charge under an open-source license |
The third row is narrower than OSAID 1.0 on training data: data information is required only when the training data itself is absent.[41] Stephen O'Grady of RedMonk wrote on June 2, 2026 that none of the 40 open models in RedMonk's sample met either open source AI tier, and that adoption of the definition "has been minimal".[46]
United States and international bodies
In 2025 OSI submitted comments on the White House AI Action Plan jointly with the Apereo Foundation and published a resource on AI and open source for federal and state policymakers. It is a member of the Digital Public Goods Alliance and endorsed the United Nations Open Source Principles.[17]
Developments since OSAID 1.0
| Date | Event |
|---|---|
| January 29, 2025 | OSI publishes its open weights explainer [25] |
| February 2025 | Data governance white paper with Open Future; second statement on the Llama license [39][15] |
| March 2025 | Disputed board election; GPAI Code of Practice draft amended [35][36] |
| September 15, 2025 | Maffulli's departure announced; Deborah Bryant named interim director [14] |
| January 28, 2026 | Spring 2026 board election cancelled pending governance review [19] |
| April 13, 2026 | Duane O'Brien starts as executive director [11] |
| May 29, 2026 | G7 ministers approve the Vision on AI Openness [41] |
| June 23, 2026 | Open Source AI Fellowship announced at UN Open Source Week [42] |
| September 24, 2026 | Board adopts governance reforms [18] |
| October 21 and 27, 2026 | First Open Source AI Salons, in Raleigh and Barcelona, open a review of "whether and how the OSAID should evolve" [45] |
The Open Source AI Fellowship is a two-year program run with Duke University's Sanford School of Public Policy, with launch sponsorship from Red Hat, AWS, Google, Automattic and Mozilla. Gabriel Toscano is the founding fellow. OSI said it had concluded from stakeholder feedback that it needed a program "inclusive of but broader than the OSAID", and the planned workshops include "an exploration of potential updates or revisions" to the definition.[42] OSI's annual report lists advancing the OSAID among its plans for 2026 and calls financial sustainability an equal priority.[17]
Continue reading
- Open-source AIOpen-source AI refers to artificial intelligence systems whose recipients have broad rights to use, study, modify, and share the system, together with access to the materials needed to exercise those rights.Mentioned 8 times above
- Open weightsOpen weights is the practice of publishing the trained parameter tensors of a neural network so that anyone can download, run, fine-tune, and redistribute the model without obtaining a separate inference…Mentioned 7 times above
- Open-Weight LLM License ComparisonAs of November 2026, the most widely used licenses that put no user, revenue or use conditions on commercial use of open-weight LLMs are two standard, OSI-approved ones: Apache-2.0 (Alibaba's Qwen3, most of…Listed in See also
- Apache License 2.0The Apache License, Version 2.0 is a permissive open source software license published by the Apache Software Foundation (ASF) in January 2004 and approved by the ASF board on January 21, 2004.Listed in See also
- MIT LicenseThe MIT License is a short, permissive software license. It grants broad permission to use, copy, modify, merge, publish, distribute, sublicense, and sell covered software, subject to one express preservation…Listed in See also
- OpenMDWOpenMDW (Open Model, Data and Weights License) is a permissive license agreement developed under the Linux Foundation specifically for machine learning models and their related artifacts.Listed in See also
More from See also
More to read
Topics:AI Policy & RegulationOpen Source AIResearch Organizations
References
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Open Source Initiative. "History of the Open Source Initiative". opensource.org, last modified June 3, 2025. opensource.org/history
- ^1 ^2 ^3Open Source Initiative. "The Open Source Definition" (version 1.9). opensource.org, last modified February 16, 2024. opensource.org/osd
- ^1 ^2 ^3 ^4 ^5Open Source Initiative. "Board of directors". opensource.org, accessed November 17, 2026. opensource.org/board
- ^1 ^2O'Brien, Duane. "Hello From The New Executive Director". Open Source Initiative, April 23, 2026. opensource.org/...-from-the-new-executive-director
- ^1 ^2 ^3 ^4Open Source Initiative. "The Open Source Initiative Announces the Release of the Industry's First Open Source AI Definition". opensource.org, October 28, 2024. opensource.org/...-first-open-source-ai-definition
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12Open Source Initiative. "Final Board Report" (report on the Open Source AI Definition co-design process). opensource.org. opensource.org/...final-board-report
- ^1 ^2 ^3 ^4 ^5 ^6 ^7Open Source Initiative. "The Open Source AI Definition - 1.0". opensource.org, accessed November 17, 2026. opensource.org/...open-source-ai-definition
- ^1 ^2Kuhn, Bradley M. "Open Source AI Definition Erodes the Meaning of 'Open Source'". Software Freedom Conservancy, October 31, 2024. sfconservancy.org/...-definition-osaid-erodes-foss
- ^1 ^2 ^3Maffulli, Stefano. "Meta's LLaMa license is not Open Source". Open Source Initiative, July 20, 2023. opensource.org/...ama-2-license-is-not-open-source
- ^1 ^2 ^3 ^4 ^5 ^6Wiggers, Kyle. "We finally have an 'official' definition for open source AI". TechCrunch, October 28, 2024. techcrunch.com/...al-definition-for-open-source-ai
- ^1 ^2 ^3 ^4 ^5Bryant, Deborah. "Welcoming the New Executive Director!" (Members Newsletter, April 2026). Open Source Initiative, April 15, 2026. opensource.org/...oming-the-new-executive-director
- ^1 ^2 ^3Open Source Initiative. "Open Source Initiative Appoints Duane O'Brien as Executive Director". opensource.org, April 15, 2026. opensource.org/...ane-obrien-as-executive-director
- ^1 ^2 ^3Open Source Initiative. "Open Source Initiative names first Executive Director". opensource.org, September 8, 2021. opensource.org/...osi-executive-director
- ^1 ^2 ^3 ^4 ^5 ^6Open Source Initiative. "OSI charts next phase for the organization with executive director search". opensource.org, September 15, 2025. opensource.org/...-phase-executive-director-search
- ^1 ^2 ^3 ^4 ^5Maris, Jordan. "Meta's LLaMa license is still not Open Source". Open Source Initiative, February 18, 2025. opensource.org/...license-is-still-not-open-source
- ^Open Source Initiative. "The License Review process". opensource.org, last modified March 13, 2024. opensource.org/...review-process
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9Open Source Initiative. "2025 Annual Report". opensource.org, 2026. opensource.org/2025-annual-report
- ^1 ^2 ^3 ^4Open Source Initiative. "2026-09-24 Board Meeting" (minutes). opensource.org, published November 13, 2026. opensource.org/...2026-09-24
- ^1 ^2 ^3 ^4Bryant, Deb. "2026 OSI Elections Update". Open Source Initiative, January 28, 2026. opensource.org/...2026-osi-elections-update
- ^Open Source Initiative. "2026-02-20 Board Meeting" (minutes). opensource.org. opensource.org/...2026-02-20
- ^Open Source Initiative. "Our supporters". opensource.org, accessed November 17, 2026. opensource.org/sponsors
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8Brockmeier, Joe. "OSI readies controversial Open AI definition". LWN.net, October 25, 2024. lwn.net/...995159
- ^1 ^2Open Source Initiative. "The Open Source AI Definition RC1 is available for comments". opensource.org, October 2, 2024. opensource.org/...-0-rc1-is-available-for-comments
- ^1 ^2 ^3 ^4Open Source Initiative. "OSAID FAQs: Answers to Frequently Asked Questions". opensource.org, accessed November 17, 2026. opensource.org/...faq
- ^1 ^2 ^3Open Source Initiative. "Open Weights: not quite what you've been told". opensource.org, January 29, 2025. opensource.org/...open-weights
- ^1 ^2Open Source Initiative. "Report from OSS EU 2025 and AI_dev: What's next for OSAID". opensource.org, September 10, 2025. opensource.org/...-and-ai_dev-whats-next-for-osaid
- ^Toscano, Gabriel. "The AI Era Arcs Toward Openness". Open Source Initiative, August 12, 2026. opensource.org/...the-ai-era-arcs-toward-openness
- ^Vidal, Nick. "2024 end-of-year review: Open Source AI Definition v1.0". Open Source Initiative, December 17, 2024. opensource.org/...w-open-source-ai-definition-v1-0
- ^Free Software Foundation. "FSF is working on freedom in machine learning applications". fsf.org, October 22, 2024. fsf.org/...reedom-in-machine-learning-applications
- ^Debian Project. "General Resolution: Interpretation of DFSG on Artificial Intelligence (AI) Models". debian.org, 2025. debian.org/...vote_002
- ^Brockmeier, Joe. "Debian debates AI models and the DFSG". LWN.net, April 25, 2025. lwn.net/...1018497
- ^1 ^2Fioretti, Marco. "Avoiding OSAID: Brock and Perens Reflect on a Year of Open Source AI Debate". FOSS Force, September 30, 2025. fossforce.com/...n-a-year-of-open-source-ai-debate
- ^Open Source Initiative. "Members Newsletter - November 2024". opensource.org, November 2024. opensource.org/...members-newsletter-november-2024
- ^1 ^2Brockmeier, Joe. "OSI election ends with unsatisfying results". LWN.net, March 21, 2025. lwn.net/...1014603
- ^1 ^2 ^3Open Source Initiative. "Announcing the new directors of OSI board". opensource.org, March 21, 2025. opensource.org/...g-the-new-directors-of-osi-board
- ^1 ^2Maris, Jordan. "Ensuring Open Source AI thrives under the EU's new AI rules". Open Source Initiative, March 27, 2025. opensource.org/...rives-under-the-eus-new-ai-rules
- ^Sippel, Birgit; Gregorova, Marketa; and others. "Mislabelling of Open Source AI and consequences for enforcement of the AI act" (letter to Executive Vice-President Henna Virkkunen). April 10, 2025. opensource.org/...commissionletterai.pdf
- ^Future of Life Institute. "Overview of Guidelines for GPAI Models". artificialintelligenceact.eu, July 30, 2025. artificialintelligenceact.eu/...uidelines-overview
- ^1 ^2Tarkowski, Alek. "Data Governance in Open Source AI". Open Source Initiative and Open Future, February 3, 2025. opensource.org/data-governance-open-source-ai
- ^Bryant, Deborah. "A Season of Change at OSI" (Members Newsletter, October 2025). Open Source Initiative, October 8, 2025. opensource.org/...a-season-of-change-at-osi
- ^1 ^2 ^3 ^4Maris, Jordan. "Open Source Initiative Helps G7 Deliver Vision On AI Openness". Open Source Initiative, June 1, 2026. opensource.org/...g7-deliver-vision-on-ai-openness
- ^1 ^2Open Source Initiative. "Open Source AI Fellowship Announced at UN Open Source Week". opensource.org, June 23, 2026. opensource.org/...announced-at-un-open-source-week
- ^O'Brien, Duane. "Opening New Paths for Open Source" (Members Newsletter, October 2026). Open Source Initiative, October 19, 2026. opensource.org/...opening-new-paths-for-open-source
- ^G7 Digital and Technology Ministers. "G7 Vision on AI Openness Opportunities and Shared Language". Paris, May 29, 2026. G7 Information Centre, University of Toronto. g7.utoronto.ca/...260529-vision
- ^1 ^2Toscano, Gabriel. "Open Source AI Salons: Shaping the OSAID, Together". Open Source Initiative, November 10, 2026. opensource.org/...alons-shaping-the-osaid-together
- ^O'Grady, Stephen. "The G7 on Open Source vs Open Weights". RedMonk tecosystems, June 2, 2026. redmonk.com/...g7-open-weights
- ^Open Source Initiative. "The Open Source AI Definition - 1.0-RC2". opensource.org, October 21, 2024. opensource.org/...pen-source-ai-definition-1-0-rc2
Improve this article
Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.
1 revision · v1 · 5,642 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Reviewer note: Independent fact-check on November 17, 2026: 128 claims checked against OSI pages, LWN and other cited sources; nine corrections applied before publication, including a misattributed remark and the G7 label table.
Cite this page: AI Wiki. "Open Source Initiative." aiwiki.ai, updated 17 Nov 2026, fact-checked 17 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/open_source_initiative